卷积混合信号频域盲分离技术研究

阵列声纳观测的舰船辐射噪声信号是由多目标源、海洋环境噪声等经多途卷积混叠形成，为提高被动声纳的检测能力、有效分离多目标信号与环境噪声，提出一种新的信号盲分离方法，利用滑窗短时傅里叶变换将时域信号卷积混合形式转换到频域瞬态混合形式，针对每个频率点利用非平稳水声信号的多拍交叉相关序列，建立频域适用盲分离算法，估计分离网络矩阵，分离恢复多源信号，研究了多源多途水声信号盲分离技术，用仿真信号和水池实验实录信号进行频域盲分离算法检验，结果表明信号分离效果较好。

Research on blind separation in frequency domain for convolutive mixed signals

Ship radiant noises obtained by an array sonar are convolutive mixtures for multiple-sources and ocean noises through multiple-paths. A novel approach of the blind source separation is proposed to improve the detection power of passive sonar and effectively separate multi-target sources with environment noises. The convolutive model in time domain is transformed into an instantaneous model in frequency domain by sliding window SFT (short FT). The multiple-time cross-correlation sequence of the non-stationary underwater acoustic signals is employed for every frequency bin, so as to establish suitable blind separation algorithm in frequency domain, estimate the separation network matrix and recover multiple-sources. The blind separation technique is studied for multiple-sources and multiple-paths underwater acoustics signals. Simulation results illustrate a good performance under both conditions of synthesized signals and recorded signals from the water-tank experiment.